Analyzing investor sentiments correctly plays a crucial role in preventing potential market risks. However, in the selection of data, existing studies predominantly focus on textual data presented by investors, while overlooking the supplementary role of other financial data in sentiment analysis. For instance, related background news, as well as responses from company secretaries to investor questions on investor interactive platforms. This oversight results in discrepancies between the analyzed sentiments and the actual sentiments. In this paper, we propose an approach that combines GPT and RoBERTa for investor sentiment analysis on Chinese investor interactive platforms, utilizing relevant news, investor questions, and secretary responses as collective analytical data to achieve accurate analysis of investor sentiments. Specifically, to uncover latent information between different texts, we employ GPT for text comprehension and fusion. Addressing instances where questions involve multiple companies in the fusion text, we emloy prompt learning with RoBERTa to conduct sentiment analysis tailored to the questioning company. The accuracy of sentiment analysis reaches 97.61%, demonstrating an improvement of approximately 10% compared to the baseline.

This study examines credit spreads on municipal bonds through an innovative cross-application approach. Based on the exacerbated economic uncertainty and the unique position of municipal bonds in the Chinese credit market, the study of municipal bonds’ credit spread prediction is necessary. In this study, we propose a hybrid RF-DWT-SAM-GRU model, and we find that the random forest (RF) for feature selection enhances the model performance well, while the discrete wavelet transform (DWT) for time-series decomposition is more suitable for this study’s dataset compared to CEEMDAN (Complete Systematic Empirical Modal Decomposition of Adaptive Noise), and the gated recurrent unit (GRU), which is augmented by the self-attention mechanism (SAM), is more accurate in captures the fluctuations and trends in credit spreads of municipal investment bonds, thus improving the prediction accuracy. By comparing the other models, this paper finds that the hybrid model has higher forecasting accuracy in the prediction of municipal bonds.

Market making (MM) is an important research topic in quantitative finance, the agent needs to continuously optimize ask and bid quotes to provide liquidity and make profits. The limit order book (LOB) contains information on all active limit orders, which is an essential basis for decision-making. The modeling of evolving, high-dimensional and low signal-to-noise ratio LOB data is a critical challenge. Traditional MM strategy relied on strong assumptions such as price process, order arrival process etc. Previous reinforcement learning (RL) works handcrafted market features, which is insufficient to represent the market. This paper proposes a RL agent for market making with LOB data. We leverage a neural network with convolutional filters and attention mechanism (Attn-LOB) for feature extraction from LOB. We design a new continuous action space and a hybrid reward function for the MM task. Finally, we conduct comprehensive experiments on latency and interpretability, showing that our agent has good applicability.

Market making is an important research topic in quantitative finance. Market makers need to continuously optimize their ask and bid prices to provide liquidity and make profits, which can be viewed as a continuous control problem. Reinforcement learning is a common method for solving sequential decision-making problems, in which an agent learns from reward signals through interactions with the environment to maximize the cumulative return. However, traditional online reinforcement learning methods can be inefficient in practice as they require the agent to interact with the environment to collect training data, which could be unstable. Additionally, exploration in financial trading can be very expensive. To address these issues, we apply offline reinforcement learning methods which use historical experience to train agents. In this paper, we present ORL4MM (Offline Reinforcement Learning for Market Making), a novel market making agent using offline training and online fine-tuning to mitigate potential losses and instabilities. We demonstrate the effectiveness of our method through experiments, where our agent outperforms all baseline models, including traditional models and online RL agents. To the best of our knowledge, we are the first to explore the application of offline reinforcement learning in market-making tasks, and we provide valuable practical experience for the deployment of reinforcement learning in financial scenarios.

Optimizing portfolios is an important concern for all investors. Nowadays, deep learning has been applied to the study of portfolio investment. Still, the widely used deep learning methods based on asset return prediction do not guarantee to maximize the performance of a portfolio. In this paper, we design a neural network with the overall return-risk ratio of the portfolio as the optimization objective to determine the optimal allocation weights of the portfolio. We design the network architecture based on the Conv-Transformer with graph attention mechanisms(CTG) to better model the temporal dependence of assets and the correlation relationship between assets. The empirical results on the Chinese stock market show that our approach achieves the best results compared to the current SOTA model.

In the Internet era, due to the rapid development of investors communication with public companies, people have diversified ways to express their opinions, thus generating a large amount of data, which contains valuable information. In this paper, we use a combination of the financial sentiment dictionary and Bert to analyze the sentiment of investors’ questions based on the Q&R data of board secretaries on the platform "Easy Interactive" (http://irm.cninfo.com.cn/) launched by Shenzhen Stock Exchange, and the final accuracy rate is 92%, which is 16% higher than the traditional sentiment analysis methods. Compared with offline research, financial news, stock forums, social software, and other data, the Q&R data selected in this paper has less noise and is more intuitive. Moreover, this paper considers knowledge in the financial domain in sentiment analysis and has domain friendliness and model generalization in the financial domain by combining the financial domain sentiment lexicon with the Bert model with adversarial training.

This paper investigates how to use deep learning methods to combine with traditional multi-factor models and construct a quantitative trading model based on an AutoEncoder algorithm (AE) to classify cryptocurrencies since 2009, so as to screen out ones with investment value and then construct an effective investment portfolio. The AE algorithm is capable of handling high-dimensional data and mining interfactor non-linearities. Our empirical results on cryptocurrencies show that the model outperforms single-type factors and benchmark in terms of Cumulative Returns and the Sharpe Ratio.

Genetic Algorithm(GA) jumps out of the traditional quantitative factor construction methods. It is a "formula first and logic later" method and makes drastic improvement for existing factors with the help of biological evolution. In this research, a large number of commonly used quantitative factors are introduced as the GA operators, and we use the factors’ Sharpe Ratio and correlation with existing factors to modify the fitness function, so as to construct the GA more suitable for financial investment system. This research has carried out a large number of variations on 206 transaction-data factors that have been used for investment. Multiple rounds of evolutionary iterations show that our research can make existing quantitative factors jump out of local optimal, find more excellent and different factors, reduce the correlation among factors, approach the truth of market data distribution constantly.

The stock price’s highly unstable fluctuation pattern makes learning efficient representation challenging to model the stock movement. The common deep learning often overfits after a few epochs of training and performs poorly in the validation set because the optimization objective is insufficient to characterize the stock adequately. In this paper, we propose Self-FTS, a self-supervised learning framework for financial time series representation, to learn the underlying representation and use in stock trading, affected by the fact that self-supervised learning is a promising technique for learning representation for extracting high dimensional features from unlabeled financial data to overcome the bias caused by handcrafted features. Specifically, we design several auxiliary tasks to generate samples with pseudo labels from the A-share stock price data sets and build a weight-sharing feature extraction backbone combined with a classification head to learn the pseudo labels based on the samples. Finally, We evaluate the learned representations extracted from the backbone by fine-tuning data sets labelled with stock returns to build an investment portfolio. Experimental analysis results on the Chinese stock market data show that our method significantly improves the stock trend forecasting performances and the actual investment income through backtesting compared to the current SOTA method, which strongly demonstrates our effective approach.

Topic modeling extracts useful potential topics that reflect market information from massive financial news and is widely used in data mining and economic research. Traditional topic modeling approaches such as Latent Dirichlet Allocation (LDA) and Non-Negative Matrix Factorization (NMF) lack semantic information, and short texts have feature sparse problems. We develop a topic clustering model based on BERT-LDA joint embedding that takes both contextual semantics and thematic narrative into account. We cluster document embeddings with the HDBSCAN algorithm and utilize a class-based TF-IDF (c-TF-IDF) method to create topic representations. Empirical results show that the BERT-LDA model is competitive compared with traditional and single topic models. It generates coherent topic words that are dissimilar to each other.

The correlation between stocks is important for investment portfolio pricing and evaluation, risk management, and formulating trading and hedging strategies. The COVID-19 has led to a general increase in the degree of correlation between stocks, the market-wide allocation has lost its meaning, and the hedging strategy has failed. It is more necessary and urgent to predict the correlation between stocks under the influence of the epidemic. However, previous studies mostly focused on traditional financial models. There are problems such as too many assumptions and restrictions, the dimensional disaster of the estimated parameters, and the poor effect of fitting nonlinearity and tail risk, which cannot provide reliable and accurate estimates. In this paper, the covariance matrix for stock return is considered as a sequence with both time and space characteristics, to transform the problem into the study of spatiotemporal sequence prediction. We Innovatively apply the end-to-end Convolutional LSTM (ConvLSTM) to the correlation prediction between stocks and use random matrix theory (RMT) to improve mean squared error (MSE) to eliminate the influence of noise. Experiments show that the performance of ConvLSTM on this problem is better than that of traditional financial model, especially after de-nosing by Random Matrix Theory (RMT). Compared with Fully Connected LSTM (FC-LSTM), ConvLSTM acquired a better out-of-sample MSE and RMT_MSE, which proves the effectiveness of the method. Finally, we repeat experiments with other stock dataset to verify the robustness of the model.

U.S. Treasury yield rates are the most important reference for global asset pricing and usually affect the stock market. Therefore, research on the correlation between China's core asset valuation and Treasury yield rates is becoming more and more important. The current statistical measurement methods have shortcomings such as the short period of market variables, low frequency, and inability to observe indicators of different countries in real-time. News, as information that reflects the public's attention and cognition, directly affects investors' stock trading behavior in the short term and has timeliness. We construct Correlation Strength by News (CSN) index for the first time to measure the correlation strength between treasury yield rates and the stock market from the perspective of media attention. The proposed method effectively solves the problem of the traditional method, such as the lack of data update timeliness and forecasting effectiveness. The capability of the index as an alternative variable of the correlation degree between the treasury yield rates and the stock market is verified.

One of the major characteristics of financial time series is that they contain a large amount of nonstationary noise, which is challenging for deep neural networks. People normally use various features to address this problem. However, the performance of these features depends on the choice of hyper-parameters. In this paper, we propose to use neural networks to represent these indicators and train a large network constructed of smaller networks as feature layers to fine-tune the prior knowledge represented by the indicators. During back propagation, prior knowledge is transferred from human logic to machine logic via gradient descent. Prior knowledge is the neural network's deep belief and teaches the network to not be affected by non-stationary noise. Moreover, co-distillation is applied to distill the structure into a much smaller size to reduce redundant features and the risk of overfitting. In addition, the decisions of the smaller networks in terms of gradient descent are more robust and cautious than those of large networks. In numerical experiments, we find that our algorithm is faster and more accurate than traditional methods on real financial datasets. We also conduct experiments to verify the method.

Algorithmic trading has become the major trading mechanism and one of the core technologies of electronic transactions globally. In USA, above 90% of electronic trading volumes has been done by algorithmic trading systems. However, algorithmic trading is still new in China capital market, only less than 10% of the volume has been done by algorithmic trading systems. With the rapid development of Chinese capital market and QFII capacity expansion, it will be the major trading mechanism in China. While being introduced into Chinese markets, it has to adapt to some special local trading rules, such as: Price limits (limit up and limit down). Because of the particular preferences by the Chinese investors, the market has a unique morphology forms in price limits. How to improve the model of price limits in China's algorithmic trading is the main focus of this research, especially under recent increasing volatility of global stock market in early 2020. This paper proposes a novel volume ratio prediction model, which can obtain a more accurate value of the price limit trading volume distribution. And an improved algorithmic trading logic based this model is proposed and proves its effectiveness.

Financial news plays an important role in investors' decisions and then influences stock markets. Previous studies mainly focus on establishing sentiment index from financial text and then making stock return prediction and trading strategy based on the index. This procedure demands costly manual label and may not directly correspond to actual stock market reaction. This paper solves this problem by using labels of stocks' residual return as sentiment labels for BERT model training. Distinct from ordinary task, buying or selling action will be taken after judgement of the stock news' sentiment. Hence, weighted cross-entropy loss and cost-sensitive accuracy are used to reveal influence and cost of judgement. Different settings of weighted cross-entropy loss are applied to learn self-adaptively and a selection method is designed to seek capable base classifiers for ensemble learning. This paper then develops a stock trading strategy based on the ensemble BERT model. Experiments and ablation study show the robust effectiveness of our strategy.

Stock return prediction can help investors make better investment decisions and trends of country's economics. However, most of methods for stock return prediction are based on time-series models, treating the stocks as independent from each other. Inter-relations among stocks' time series are out of consideration. In this work, a Multi-Channel Temporal Graph Convolutional Neural Network (MCT-GCN) is proposed to optimize stock movement prediction. Experiments show that its performance is greater than benchmark algorithms, LSTM in the S&P 500.

To extend the application of a previously developed home ECG and blood pressure telemonitoring system for pacemaker users, a specially designed pacemaker working status analysis algorithm is developed in this paper. This paper includes four sections. First, a pacing beat and spontaneous beat recognition algorithm are established in which a FIR filter and three threshold methods are used to pick up the pacing spikes and a dynamic threshold modification method is proposed for error elimination. Second, a beat classification algorithm is developed based on decision rules for six commonly used pacemaker types. Third, a pacemaker malfunction recognition algorithm is established to evaluate the pacemaker working status. Fourth, the statistics of the analysis and the report are also provided. To validate the algorithm, clinical data were collected and manually classified by specialists. Then, the classification and recognition results were obtained with the clinical data. Results show that the algorithm can achieve a recognition rate of 98.4%. Results also indicate that the algorithm is capable of working in real-time speed. The algorithm has been incorporated into the telemonitoring system, and trial application cases are also reported.